Bug 313494 - [ sensor / dynamics ] Hue / Saturation / Value sensors are broken
Summary: [ sensor / dynamics ] Hue / Saturation / Value sensors are broken
Status: RESOLVED FIXED
Alias: None
Product: krita
Classification: Applications
Component: Brush engines (show other bugs)
Version: git master (please specify the git hash!)
Platform: Other Linux
: NOR normal
Target Milestone: ---
Assignee: LukasT
URL:
Keywords:
: 321433 324876 325808 (view as bug list)
Depends on:
Blocks:
 
Reported: 2013-01-19 10:58 UTC by David REVOY
Modified: 2014-07-05 06:54 UTC (History)
9 users (show)

See Also:
Latest Commit:
Version Fixed In:
Sentry Crash Report:


Attachments

Note You need to log in before you can comment on or make changes to this bug.
Description David REVOY 2013-01-19 10:58:29 UTC
Hue / Saturation / Value  sensors are absolute and non relative to the user color.  Witch make them 3 uninteresting sensors.
Here is a screenshot showing the 3 items I'm talking about : http://img195.imageshack.us/img195/8826/selection012.jpg

Ex :
1) User have a blue color , pixel brush preset
2) he wants to make a brush where all the dab have a variation of value , hue or saturation
3) he opens the brush-settings , check ( for exemple ) the value box
4) then he check the Fuzzy 'sensor input' to have randomness on his value
5) then he reduce the amount (top bar combo box ) to 0.15  ( the range size of how it affect his brush. 

Actual result :
The user have a brush who does black strokes, with variation from black 0% to black 15% ... Even if he have Blue as a selected color. 

Expected result :
The user have a brush who paint blue ( active color ) with a variation of value of 15% around the current blue value. ( way more interresting ) 

Note : 
Result is actually similar with Hue , Saturation. 
Having a "relative to the active color" system could make the brush more interresting : subtle hue and saturation variation for paint strokes for richer color vibration in artwork.  All other package works this way. Sometime they offer a check box to switch absolute ... but default is relative.
Comment 1 LukasT 2013-02-16 13:41:59 UTC
The problem was that UI was confusing.

If you set the curve to 50% on both ends, you get selected color for every hsv color sensor.
Check documentation here http://userbase.kde.org/File:Krita-tutorial4-IV.2-1.png

Labels will be changed to make it more intuitive to use.
Comment 2 David REVOY 2013-02-16 14:48:31 UTC
http://www.davidrevoy.com/XYZ/2013-02-16_Krita-hue-saturation-value_bug_xvid.avi 
A HD video* ( 11MB ) Xvid 1600x900 to see the Bug in motion.

Description of video event : I trace a stroke on intro to show the basic Pixel Brush , with a pressure sensor affecting size only and the color I'll use. Then I go to the brush editor, and try to affect the Hue. The brush preview test area  fail with the bug ( 2 pure color ) , but the first stroke on canvas is OK ( surprise ) , then I return to the test area and fail again. So , I try other setup, changing ; and after that only failure for Saturation too ( showing RED and CYAN described on the bug report ) and also Value going from white to black without the color. 
In the test I try to switch also to openGL without success. It worked once. 
 
( * temp link, will expire in one or two month )
Comment 3 LukasT 2013-02-16 19:17:08 UTC
Git commit dae2629fa63bc0f54f1cc98a9fe4514c68c335cd by Lukáš Tvrdý.
Committed on 16/02/2013 at 20:12.
Pushed by lukast into branch 'master'.

HSV sensors: Improve UI intuitivness

Add labels to explain the change of color characteristics

o CW hue: Clock-wise change of hue (hue in hsv model is circle)
o CCW hue: Counter-clock wise change
o Lower/Higher value
o More/Less saturation

M  +30   -2    krita/plugins/paintops/libpaintop/kis_pressure_hsv_option.cpp

http://commits.kde.org/calligra/dae2629fa63bc0f54f1cc98a9fe4514c68c335cd
Comment 4 LukasT 2013-02-16 19:24:25 UTC
I can't reproduce that problem on the video...
Can you try on different system or with clean build of master or can you ask some Krita users to try to reproduce?
Comment 5 LukasT 2013-02-18 17:14:39 UTC
I tried to reproduce the problem in deevad's video on 32bit arch / intel graphics / Core i7
Not reproducible.
Comment 6 honoomahoujutsushi 2013-02-18 17:17:45 UTC
Reproduction attempted on:

- 64-bit Kubuntu 12.10 with Kubuntu-Backports PPA packages (including Krita 2.6)
- ATI Radeon graphics using the xorg-video-radeon drivers
- Intel Core i7-740QM 1.73GHz

Outcome: Unable to reproduce the bug.
Comment 7 David REVOY 2013-02-18 17:24:02 UTC
- 64-bit Linux Mint KDE 12.10 
- Nvidia 310 proprietary
- Intel Core i7-870 2.93GHz

Vc is a requirement on my PC to build Krita, it won't build with a version older than master 7 february. 
Bug reproduced ( Clean install VC / clean install Krita from yesterday. )
Comment 8 Halla Rempt 2013-05-18 09:12:13 UTC
Btw, for Vc, you should use the 0.7 branch, not master.
Comment 9 Paul Geraskin 2013-06-04 07:28:23 UTC
Sorry for bothering. But Scatter parameter is also in absolute values(pixels). But it should be relative to brush size.
Here is my bug report with a screenshot: https://bugs.kde.org/show_bug.cgi?id=320488
Comment 10 Matita2B 2013-06-08 08:35:22 UTC
I'd like to report I have the same brush behaviour shown by David Revoy's video in comment #2:

- "Value" seems to completely ignore the hue component of active colour and paints only a scale of grey
- "Hue" seems to go only from the active colour to neutral grey, without anything inbetween.

Some system specification:
- Linux Mint 15 MATE on VirtualBox 4.2.12 (OSX 10.6.8, MacBook5,2, 2 GHz Intel Core 2 Duo, Nvidia GeForce 9400M)
- Open source graphic drivers
- Krita 2.8 prealpha installed from Krita Lime PPA, last updated 2013-06-07, OpenGL option inactive

I checked a prior version on a different system and the brush shows the expected behaviour (as in documentation LukasT linked in comment #1):
- Linux Kubuntu Raring on VirtualBox (same as above)
- Open source graphic drivers
- Krita 2.6.3 installed from Kubuntu repositories, OpenGL option inactive

Let me know if I have to give other informations, or if there's some other step to check I'm not aware of.
Comment 11 Halla Rempt 2013-06-20 16:27:51 UTC
*** Bug 321433 has been marked as a duplicate of this bug. ***
Comment 12 Halla Rempt 2013-06-20 16:28:19 UTC
Two artists confirming the issues seems to me a proper confirmation indeed.
Comment 13 Halla Rempt 2013-09-28 08:19:20 UTC
*** Bug 324876 has been marked as a duplicate of this bug. ***
Comment 14 Halla Rempt 2013-10-12 09:31:15 UTC
*** Bug 325808 has been marked as a duplicate of this bug. ***
Comment 15 Spencer Brown 2014-03-17 22:43:03 UTC
I can reproduce this on my machine against Vc 0.7.3 from the Arch Linux repositories. I'm going to take a crack at fixing it.
Comment 16 Halla Rempt 2014-03-19 09:29:04 UTC
Awesome!
Comment 17 wolthera 2014-06-19 22:42:16 UTC
Alright, this evening I spend trying to figure out this bug. What I've found:

1. None of the three parameters adjust the current colour like they are supossed to. Instead, they overwrite them.
2. In this, the saturation is always set to 0 if not explicitely defined by the saturation sensor.
2b, so what you get is that you turn on Hue, but only paints grey because the saturation is 0.
3. Because the sensors are set up so that they can modify in both directions, we get wonderful things like Hue being -0.6, which doesn't work, so you only get the default value.
3b. Similarly, negative saturation in this case leads to the complementary colour.
4. Value is not Value. It's lightness. And, it's also the only one adjusting based on the current colour.

Which made me make a connection:
http://quickgit.kde.org/?p=calligra.git&a=commit&h=12737c6b353db80ca75a3ab819a5de3764a0e963

My suspicion is that these sensors are trying to adjust HSL coordinates while thinking they are HSV coordinates. However, I don't really understand pigment, and don't have a clue where to go from here.
Comment 18 wolthera 2014-06-20 21:38:25 UTC
This bug makes me sad.

After scourging pigment for the used options(namely kis_hsv_adjustments) I come to the conclusion it's not there.

My next stop was kis_brushop.cpp, lines 130 to 135.
There, we see the hsv options being put into a transform. Consequently, the transforms are applies to a color source, m_colorsource.

This reffers to kis_color_source line 76, applyColorTransformation
There, you see the hsv transformation being done, it has three parameters:
source pixel, destination pixel and nPixels, the later meaning how often this transformation is itterated(?).
Anyhow, the last parameter is set to 1. When you set it to 0 the HSV options don't work at all, because duh, there's 0 itterations of the color transformation being done.

So, I debugged the pixel being input with the following:
qDebug()<<m_color->toQColor();

This showed that regularly, if you had H, S and V active, the pixel color being input was black.
On other places in the file, it showed that m_color was being create by, amongst others, the mix parameter. So turning that on(while deactivating the sensors), m_color started returning a color other than black.

It even started updating said pixel.

However.
It didn't update the color being painted.

Which leaves us with the confusing situation where:
*We have found the right color transformation, deactivating it turn off the results.
*The pixel transformed at HSV adjustment is NOT m_color, despite it being m_color that is specified.
*The color is also sent to the active foreground color, while not being m_color.

Another weird thing: Value is not similar to lightness. No, activate hue and saturation, but turn off sensor input: Now paint with yellow, and then turn the hue to blue: The color becomes much darker, even though HSV wise it shouldn't change at all.

This bug is breaking me.